Xray Imaging Spectrometer (XIS)


XIS Info Home News About People Gallery Calibration Monitoring Team Page XIS Collaborators ISAS/JAXA Kyoto U. Osaka U. Suzaku Info GO Facility Suzaku at ISAS XRS at GSFC HXD at Tokyo U. XRT at GSFC 
BackgroundIn late November 2005, it became apparent that the lowenergy effective of each XIS instrument area was rapidly decreasing. We have since concluded that this is caused by a buildup of molecular contamination on the optical block filters (OBFs) which sit just above the CCD chips in the light path of each XIS. The amount and rate of contamination is different for each instrument, and it appears to be spatially nonuniform. Our best model suggests it is composed of primarily carbon and oxygen. This page details monitoring of the onaxis contamination rate by the MIT XIS team. Included are presentation summaries as well as records of updates to the contamination model and trends. For current information applicable to GO data analysis, please refer to the Suzaku ABC Guide. 2015 January 29XIS contamination updateOptical depth and total effective area at two energies. Dashed lines in the top panel are CALDB contami_20140825. Dashed lines in the bottom panel are extrapolations from recent data. Download single PDFA new CALDB file was constructed (20140825). XIS0 and XIS3 needed slight adjustments to the O column density at recent times, while XIS1 needs an adjustment to the N trend based on PKS2155 and RXJ1856, and then a refit of the E0102 data, which freezes the N column based on that trend. Download single PDFLine center offsets of the bright E0102 emission lines, showing how the gain correction is doing. Not well. Download single PDF2014 July 1XIS contamination updateThe latest E0102 observation, 20140421, has been added to the contamination trend plot. In the plots of column density below, the red line is the current CALDB (contami_20130813) and the stars are spline points used to create that trend. A new CALDB file will be constructed from these data this month. XIS0 and XIS3 need slight adjustments to the O column density at recent times, while XIS1 needs an adjustment to the N trend based on PKS2155 and RXJ1856, and then a refit of the E0102 data, which uses freezes the N column based on that trend. Download single PDFOptical depth and total effective area at two energies. Dashed lines in the top panel are CALDB contami_20130813. Dashed lines in the bottom panel are extrapolations from recent data. Download single PDFLine center offsets of the bright E0102 emission lines, showing how the gain correction is doing. Not well. Download single PDF2014 April 1XIS contamination updateThe latest E0102 observation, 20140316, has been added to the contamination trend plot. In the plots of column density below, the black line is the current CALDB (contami_20130813). Spectral fits are also shown. This is the first E0102 observation taken after the Jan 2014 UVC problem, and there is no obvious change in the lowenergy spectra or the trend of the contamination, which still appears to be decreasing. Download single PDF2013 December 5XIS contamination updateThe latest E0102 observation, 20130929, has been added to the contamination trend plot. In the plots of column density below, the black line is the current CALDB (20130705). Download single PDF2013 July 5XIS contamination updateThe contamination trend since ~ 2012 has been updated. In the plots of column density below, the stars are cubic spline points and the solid lines are interpolations of the trends. Red is the current CALDB (20120719), black is the update (20130705). This should be implemented in the CALDB by the end of August. The plan is to update this every 6 months from here on: Tsujimotosan and Wadasan will send me values for RXJ1856 and PKS2155, and I will add E0102 and fit the spline. Note the divergence in C and O for XIS1 in 20122013. The reason is unclear. Download single PDF2013 May 24XIS contamination by element (H, C, N, O) for each XISDownload single PDFXIS contamination by XISDownload single PDFE0102 spectral fits(using EDM's nifty gain correction algorithm)Download single 170page PDF 2011 October 4Using the same fits as 20110927, here are the FWHM and gain trends, along with (for completeness) the contamination history with the spline fits. PDFHere are spectra from 20110611, the most recent observation, showing the differences between the current CALDB contamination model (ae_xi*_contami_20091201.fits) and the spline fit detailed in 20110927. single PDF file of all 3 detectors2011 September 27All E0102 data up till 20110629 are now included, using only normal fullwindow mode. All XIS1 SCI=2 keV and SCI=6 keV data are included. The bestfit trends are from proposed cubic spline interpolation, since simple analytic functions do a poor job of fitting the trends at all epochs. XIS1 and XIS2 are monotonically increasing. XIS0 also monotonically increases, but there are inflection points where the concavity changes (seen more easily in the C column history, below). XIS3 as fit is not monotonically increasing; this appears warranted in the C column history below, but it should be discussed, as should the use of byeye cubic spline fitting. Attached below are CALDB contamination files, in the same format as ae_xi*_contami_20091201.fits, with updated C,O,H columns reflecting these trends. The O/C and H/C ratios remain as in the previous HCO model. The cubic spline has been extrapolated to 2015. Also attached are data files of the cubic spline points for each XIS. O column history  PDFC column history  PDF Updated contamination files: ae_xi0_contami_edm_20110927.fits ae_xi1_contami_edm_20110927.fits ae_xi2_contami_edm_20110927.fits ae_xi3_contami_edm_20110927.fits Spline points (columns are MJD and C column density in 10^{18} cm^{2}): splinepoints_xis0.dat splinepoints_xis1.dat splinepoints_xis2.dat splinepoints_xis3.dat 2011 August 22All E0102 data up till 20110629 are now included, using only normal fullwindow mode. All XIS1 SCI=2 keV and SCI=6 keV data are included. The updated CALDB contamination trends from ae_xi*_contami_20091201.fits are plotted; these use the timevariable HCO model. Also changed from previous plots: the O column density is now being plotted instead of C; and the errorbars are 1sigma instead of 90%. PDF2010 October 06
New E0102 data from 2010 August 29, in normal fullwindow mode
for XIS0 and XIS1. XIS3 was in PSum mode, and is excluded.
First, the bestfit contamination assuming C/O = 6 and
with trends from the current CALDB files: Here are the gain offsets and excess linewidth for the same CALDB fit. "Gain offset" is an additive energy scale offset from the XSPEC "gain fit" command, positive gain shift means that the PI values are higher than the model, so the model line energies must be shifted up. Open points indicate SCIon data, crosses indicate SCIoff data. PDF2010 August 13
New E0102 data from 2010 June 19, in normal fullwindow mode
for all three chips. First, the bestfit contamination assuming C/O = 6 and
with trends from the current CALDB files: Here are the gain offsets and excess linewidth for the same CALDB fit. "Gain offset" is an additive energy scale offset from the XSPEC "gain fit" command, positive gain shift means that the PI values are higher than the model, so the model line energies must be shifted up. Open points indicate SCIon data, crosses indicate SCIoff data. PDFFinally a spectrum from XIS1 of the latest observation (2010 June 19) showing an excess of counts below 0.5 keV compared to the E0102 emission model plus C+O contamination model: PDF2010 July 22No contamination update, but some investigation into gain errors at low energy. Below are the bestfit line centers for the OVII, OVIII, NeIX, and NeX from all E0102 observations on XIS0 and XIS1, both SCI off and on, using the most current calibration products to date. A single Gaussian was fit to each line along with a Bremsstrahlung continuum and contamination fixed to the bestfit value for that date. The plotted lines are simple unweighted leastsquares fits to all the data points, and sigma_y is the scatter about this fit. Errors are 1sigma. There is a clear trend of increasing line center in XIS1 for all four emission lines, and slightly less in XIS0, perhaps due to overcorrection of CTI. The first two data points for XIS1 in 2005 are about 5 eV higher than the trend would predict. This is probably not a difference due to SCI, as more than half of the datasets in 2006 are also SCIoff. PDF (all figures in one file)2010 May 25
New E0102 data from 2010Feb05 and 2010Apr05, in normal fullwindow mode
for all three chips. First, the bestfit contamination assuming C/O = 6 and
with trends from the current CALDB files: Here are the gain offsets and excess linewidth for the same CALDB fit. "Gain offset" is an additive energy scale offset from the XSPEC "gain fit" command, positive gain shift means that the PI values are higher than the model, so the model line energies must be shifted up. Open points indicate SCIon data, crosses indicate SCIoff data. PDF2010 Jan 16
New E0102 data from 2009Dec12 for XIS1 only; XIS0,3 were observed in
PSum mode and so are not included. Here are the gain offsets and excess linewidth for the same CALDB fit. "Gain offset" is an additive energy scale offset from the XSPEC "gain fit" command, positive gain shift means that the PI values are higher than the model, so the model line energies must be shifted up. Open points indicate SCIon data, crosses indicate SCIoff data. PDFNext, the bestfit contamination assuming the empirical HCO model and with trends from 2009Oct11: PDFFinally, optical depth and transmittance. The top plot shows the evolution of the optical depth at two energies, for each detector. The bottom plot shows the total effective area for all detectors summed. Each trend has been extrapolated (dashed lines) at its current rate through the end of 2012. PDF2009 Dec 21New E0102 data from 2009Oct11/12 and 2009Oct27, with fits using the HCO empirical contamination model. Attachments: Contamination rate plot PDF2009 Oct 11Analysis of HCO empirical contamination model with E0102. Please see the attached PDF file. Attachments: PDF2009 July 21Updated with the 2009 April 23 and 2009 June 26 observations. The processing and trend fits are the same as described in 20090402. This still assumes C/O = 6 and does not yet use the new E0102 emission model from IACHEC. The last datapoint (2009 June 26) occurred after the XIS0 charge leakage problem (on 2009 June 23). There are no measurable effects on the FWHM or gain after this event, indicating that the calibration is likely unaffected for quads B and C, where the source is detected. Attachments: Contamination rate plot PDFBestfit "extra" linewidth and gain offset vs. date. Linewidth measures how accurately the RMF reflects the "true" linewidth (as produced by the instrument and processing). "Gain offset" is an additive energy scale offset from the XSPEC "gain fit" command, positive gain shift means that the PI values are higher than the model, so the model line energies must be shifted up. Open points indicate SCIon data, crosses indicate SCIoff data. Linewidth trend plot PDFGain trend plot PDF 2009 May 18Updated with the 2009 March 9 observation. The processing and trend fits are the same as described in 20090402. This still assumes C/O = 6 and does not yet use the new E0102 emission model from IACHEC. Attachments: Contamination rate plot PDFBestfit "extra" linewidth and gain offset vs. date. Linewidth measures how accurately the RMF reflects the "true" linewidth (as produced by the instrument and processing). "Gain offset" is an additive energy scale offset from the XSPEC "gain fit" command, positive gain shift means that the PI values are higher than the model, so the model line energies must be shifted up. Open points indicate SCIon data, crosses indicate SCIoff data. Linewidth trend plot PDFGain trend plot PDF 2009 April 02Updated processing (gain and FWHM calibration) including all data through the 2008 Dec 14 observation. A new functional form for the contamination rate has been adopted: N_C = A (1+Ct) [1exp(t/B)] where the parameters are:
Full details of the processing and an anlysis of the temporal variations is included in this memo: Attachments: Contamination rate plot PDFBestfit "extra" linewidth and gain offset vs. date. Linewidth measures how accurately the RMF reflects the "true" linewidth (as produced by the instrument and processing). "Gain offset" is an additive energy scale offset from the XSPEC "gain fit" command, positive gain shift means that the PI values are higher than the model, so the model line energies must be shifted up. Open points indicate SCIon data, crosses indicate SCIoff data. Linewidth trend plot PDFGain trend plot PDF 2009 Jan 20Updated with the 2008 Dec 14 observation. The processing and trend fits are the same as described in 20081201. This still assumes C/O = 6 and does not yet use the new E0102 emission model from IACHEC. The contamination has held steady for XIS1 and XIS3, but continues to increase for XIS0. The solid black line for XIS0 shows the trend currently implemented in CALDB, while the dotted black line shows the 20081201 fit to data after April 2007. Attachments: Contamination rate plot PDFBestfit "extra" linewidth and gain offset vs. date. Linewidth measures how accurately the RMF reflects the "true" linewidth (as produced by the instrument and processing). "Gain offset" is an additive energy scale offset from the XSPEC "gain fit" command, positive gain shift means that the PI values are higher than the model, so the model line energies must be shifted up. Open points indicate SCIon data, crosses indicate SCIoff data. Latest four data points use RMFs generated with CALDB version rmfparam_20080311. Prior data use rmfparam_20080121. Linewidth trend plot PDFGain trend plot PDF 2008 Dec 01Updated with the 2008 Oct 22 observation. The processing and trend fits are the same as described in 20080404. This still assumes C/O = 6 and does not yet use the new E0102 emission model from IACHEC. The contamination has held steady for XIS1 and XIS3, but continues to increase for XIS0. (The last two XIS0 points are right on the XIS3 points.) I note there seems to be a correlation between the XIS0 contamination increase after April 2007 and the gain shift increase during the same period; however the latest point has a much smaller gain shift. I have not reprocessed the previous data, which all use PROCVER <= 2.2.8.20. The latest point uses PROCVER 2.2.11.32. The solid black line for XIS0 shows the trend currently implemented in CALDB, while the dotted black line shows a new fit to data after April 2007: N_C = a*(1exp((mjd0mjdbr)/b)) + c*(1exp((mjdbrMJD)/d)) for XIS0 (MJD > mjdbr) # XIS0 # broken exponential current CALDB (solid) new fit (dotted) a 3.939 a 3.939 b 270.069 b 270.069 c 1.827 c 4.007 d 238.417 d 728.928 mjd0 53595.408 mjd0 53595.408 mjdbr 54200.6 mjdbr 54200.6 The new XIS0 contamination trend has a small effect on count rate calculations for AO4. For an observation planned in Sep 2009, here are the differences in total (XIS0+1+3) effective area and count rates for the assumed CALDB trend and the new trend:
The second column shows the ratio of the total effective area on 2009 Sep 01 to the contaminationfree effective area (i.e. at launch), assuming the trends in the current CALDB (solid lines on the plot). The third column is that ratio assuming XIS0 increases according to the dotted line. The count rate at 0.5 keV will be 4% lower than calculated for the current trends. Of course, if XIS0 contamination levels off like XIS3, then the discrepancy will be smaller than this. Attachments: Contamination rate plot PDFBestfit "extra" linewidth and gain offset vs. date. Linewidth measures how accurately the RMF reflects the "true" linewidth (as produced by the instrument and processing). "Gain offset" is an additive energy scale offset from the XSPEC "gain fit" command, positive gain shift means that the PI values are higher than the model, so the model line energies must be shifted up. Open points indicate SCIon data, crosses indicate SCIoff data. Latest four data points use RMFs generated with CALDB version rmfparam_20080311. Prior data use rmfparam_20080121. Linewidth trend plot PDFGain trend plot PDF 2008 Sep 04Updated with the 2008 Aug 13 observation. The processing and trend fits are the same as described in 20080404. This still assumes C/O = 6 and does not yet use the new E0102 emission model from IACHEC. The contamination has held steady for XIS1 and XIS3, but increased again for XIS0. (The last XIS0 point is right on the XIS3 point.) I note there seems to be a correlation between the XIS0 contamination increase after April 2007 and the gain shift increase during the same period. I don't know what that mean yet. Attachments: Contamination rate plot PDFBestfit "extra" linewidth and gain offset vs. date. Linewidth measures how accurately the RMF reflects the "true" linewidth (as produced by the instrument and processing). "Gain offset" is an additive energy scale offset from the XSPEC "gain fit" command, positive gain shift means that the PI values are higher than the model, so the model line energies must be shifted up. Open points indicate SCIon data, crosses indicate SCIoff data. Latest data (20080605) uses RMFs generated with CALDB version rmfparam_20080311. Linewidth trend plot PDFGain trend plot PDF 2008 Jun 25Updated with the 2008 June 05 observation. The processing and trend fits are the same as described in 20080404. This still assumes C/O = 6 and does not yet use the new E0102 emission model from IACHEC. Attachments: Contamination rate plot PDFBestfit "extra" linewidth and gain offset vs. date. Linewidth measures how accurately the RMF reflects the "true" linewidth (as produced by the instrument and processing). "Gain offset" is an additive energy scale offset from the XSPEC "gain fit" command, positive gain shift means that the PI values are higher than the model, so the model line energies must be shifted up. Open points indicate SCIon data, crosses indicate SCIoff data. Latest data (20080605) uses RMFs generated with CALDB version rmfparam_20080311. Linewidth trend plot PDFGain trend plot PDF 2008 Apr 29Updated with the 2008 April 08 observation. The processing and trend fits are the same as described in 20080404. Attachments: Contamination rate plot PDFBestfit "extra" linewidth and gain offset vs. date. Linewidth measures how accurately the RMF reflects the "true" linewidth (as produced by the instrument and processing). "Gain offset" is an additive energy scale offset from the XSPEC "gain fit" command, positive gain shift means that the PI values are higher than the model, so the model line energies must be shifted up. Open points indicate SCIon data, crosses indicate SCIoff data. Data from Jan 2006 and later use RMFs generated with CALDB version rmfparam_20080121. Linewidth trend plot PDFGain trend plot PDF 2008 Apr 4Several updates to the contamination fitting:
In general, the contamination is slightly lower than in previous fits, but by much less than the systematic error and barely noticeable in the plot. I suspect the change is due to details of fitting the bright lines with the improved FWHM vs. the "intrinsic" linewidth hack that I use in XSPEC, but I have not confirmed this. For the contamination trend, I have not changed the exponential fits for XIS1,2,3 and XIS0 before April 2007. After April 2007, there is an altered version of the broken exponential compared to 2007 Dec 19 and 2008 Mar 3. For completeness, here are the full fitting parameters: N_C = a*(1exp((mjd0MJD)/b)) for XIS1,2,3 and for XIS0 (MJD <= mjdbr) N_C = a*(1exp((mjd0mjdbr)/b)) + c*(1exp((mjdbrMJD)/d)) for XIS0 (MJD > mjdbr) # XIS0 # broken exponential a 3.939 b 270.069 c 1.827 d 238.417 mjd0 53595.408 mjdbr 54200.6 # XIS1 # single exponential a 4.678 b 208.834 mjd0 53595.379 # XIS2 # single exponential a 4.299 b 117.472 mjd0 53595.349 # XIS3 # single exponential a 5.795 b 90.843 mjd0 53595.348 with N_C = C column density in 10^{18} cm^{2}, C/O = 6 by number, and MJD is the Mean Julian Day. mjd0 is about 53595.4, or 20070813, when the XIS doors were opened. Attachments: Contamination rate plot PDFBestfit "extra" linewidth vs. date. This measures how accurately the RMF reflects the "true" linewidth (as produced by the instrument and processing). Open points indicate SCIon data, crosses indicate SCIoff data. Data from Jan 2006 and later use RMFs generated with CALDB version rmfparam_20080121. Linewidth trend plot PDF2008 Mar 27The 2008 March 15 data have been reprocessed with the updated FWHM parameters (makepi_20080131). The bestfit contamination for XIS0 for this date has dropped by ~ 3% as a result of the new processing (from 4.67 to 4.51 x 10^{18} C atoms cm^{2}). The 90% error bar on this data point is about 1.5%, so this is a significant change. XIS1 and XIS3 contamination values also change a bit with the new processing, but not as significantly. Attached are the new rate plot (left) and the old rate plot (right, from 2008 Mar 26). Only the last data point has been reprocessed. Attachments:
Bestfit "extra" linewidth vs. date. This measures how accurately the RMF reflects the "true" linewidth (as produced by the instrument and processing). Open points indicate SCIon data, crosses indicate SCIoff data. The latest data point has been reprocessed with makepi_20080131. Linewidth trend plot PDF2008 Mar 26Updated contamination history with 2008 March 15 observation of E0102. The data for the March and Feb 2008 points were processed by EDM and are missing some of the normal processing (attitude correction, for example), as described in 2008 Feb 20 entry. This processing uses gain parameters from ae_xi*_makepi_20071031.fits, not the latest version (20080131) which has updated FWHM parameters. Also attached are a PDF file comparing spectra between Feb/Dec and Mar/Feb, including the ratio of each spectrum; and a PDF file checking for correlations between contamination and various fitting parameters (gain offset, intrinsic or "extra" linewidth, and RXJ0103 normalization). Attachments: Contamination rate plot PDFComparison spectra for Dec 2007, Feb 2008, and Mar 2008 (PDF only) Correlation plots (PDF only) 2008 Mar 3Updated fit to the recent XIS0 increase. This is the same as the 2008 Feb 20 plot, except the XIS0 postApril 2007 contamination trend is refit: XIS0 is fit with a sort of broken exponential with a break around April 2007 (MJD 54200.6). The form is: N_C = a*(1exp((mjd0MJD)/b)) for MJD <= mjdbr N_C = a*(1exp((mjd0mjdbr)/b)) + (ca)*(1exp((mjdbrMJD)/d)) for MJD > mjdbr with N_C = C column density in 1018 cm2, C/O = 6 by number, and MJD is the Mean Julian Day. mjd0 is 53595.4, or 20070813, when the XIS doors were opened. This is slightly different from the form used before for the broken power law, but only in the particulars of the parameter definition, not the functional form. The parameters are: # XIS0 # broken exponential a 3.939 b 270.069 c 5.7 d 180. mjd0 53595.408 mjdbr 54200.6 The other chip fits and that for XIS0 before April 2007 have not been changed. Attachments: Contamination rate plot PDF2008 Feb 20The attached plot (one of them) shows the XIS contamination history including the latest E0102 observation from 20080215, after the CCD warmup. The rightmost points are from that observation, and the solid lines show the previous bestfit trends. There's clearly less contamination in XIS1 and XIS3, and XIS0 is stable from the last point (or lower than it would have been if it kept on its increasing trend). The data used here were processed by me and are missing some of the normal processing (attitude correction, for example), but I don't think that's affecting things much. I've also attached a plot showing E0102 spectra from 20071202 and 20080215, and there is very little difference from one to the other. The gain and spectral resolution have not changed noticeably at these energies beyond the trends I've been seeing over the last year. Attachments: Contamination rate plot PDFComparison spectra PDF 2007 Dec 19I've attached the latest rate plot. The dotted lines show the fit through April 2007 (the current CALDB version). Solid lines show the fits listed here. Here are the new parameters for the XIS central contamination rate, assuming C/O = 6, as measured by E0102 through the 20071202 observation. Most of the data are rev2.0.6.13, and I have reprocessed them to update the PI values as per the instructions on the web page. The model is the same as before, however I allow a variable gain offset and extra linewidth linked among the 25+ Gaussian emission lines. This drastically improves the fit for SCIon data taken after early 2007, as I believe you are aware. The typical "extra" Gaussian sigma is 1520 eV in all chips, and typical gain shift is +5 eV in XIS0 and less in the other chips. (So the energy of the data are too high compared to the model in XIS0.) I will compile this information and send it on, but probably not before the holidays. XIS1,2,3 are fit with a single exponential function, as before: N_C = a*(1exp((mjd0MJD)/b)) with N_C = C column density in 1018 cm2, C/O = 6 by number, and MJD is the Mean Julian Day. mjd0 is about 53595.4, or 20070813, when the XIS doors were opened. XIS0 is fit with a sort of broken exponential with a break around April 2007 (MJD 54200.6). The form is: N_C = a*(1exp((mjd0MJD)/b)) for MJD <= mjdbr N_C = a*(1exp((mjd0mjdbr)/b)) + c*(1exp((mjdbrMJD)/d)) for MJD > mjdbr The parameters are: # XIS0 # broken exponential a 3.939 b 270.069 c 5.562 d 762.720 mjd0 53595.408 mjdbr 54200.6 # XIS1 # single exponential a 4.678 b 208.834 mjd0 53595.379 # XIS2 # single exponential a 4.299 b 117.472 mjd0 53595.349 # XIS3 # single exponential a 5.795 b 90.843 mjd0 53595.348 Attachments: Contamination rate plot PDF2007 Dec 12Here is an updated contamination history, including the E0102 observation on 2007 Dec 02. The XIS0 contamination has not increased since October. It is consistent with no change, within the uncertainty, and in fact looks a bit lower. I've included a freelyvarying gain shift in the new analysis, so the other data points have changed a bit compared to one another, but not overall. The average gain shift (for the SCIon data) ranges from +5 to +10 eV for XIS0 and XIS1, but much lower (+1 to +3) for XIS3. This likely causes a systematic error similar to the one I've shown on the plot, but not enough to explain the XIS0 contamination increase. Attachments: Contamination rate plot PDF2007 Dec 5I've analyzed the rev2 E0102 data through 20071025. This is all rev2.1.6.15 or earlier (mostly rev2.0), so I followed the instructions on this web page to correct the PI values for the SCIon data: http://heasarc.gsfc.nasa.gov/docs/suzaku/analysis/sci_gain_update.html I'm also using the latest CALDB and FTOOLS versions (as of today, at least). I enclose the updated contamination rate plot. The lines are the same as the plot from 20071001; the dotted lines are the trends from the original fit, and the solid lines are the trends fit to the 20071001 update. I have not included any fits to these points yet. The XIS0 contamination is increasing very quickly, with the latest sample almost near the XIS3 value. The enclosed spectrum (e0102_20061022_spec.pdf) verifies this: green = XIS0 20061022 black = XIS0 20071025 blue = XIS3 20061022 red = XIS3 20071025 (Sorry for the poorlyannotated plot.) The effective area has clearly decreased for XIS0 over the last year, while it has stayed roughly constant for XIS3. Contamination on XIS1 continues to increase slowly. The data points prior to April 2007 don't match the lines very well (especially for XIS0). In the prior fitting, I used a single RMF for the SCIon data (what you get from running xisrmfgen with a date of 20050813). Now I am using xisrmfgen with the full SCI CTI parameters, and the results are not very good, with poor reproduction of the redistribution function at low energies (see attached figure e0102_20061022_xis0_spec.pdf). The line centers are not bad (off by less than 10 eV), but the wings are not reproduced well. This probably accounts for much of the XIS0 discrepancy from Oct 2006 to April 2007. Attachments: Contamination rate plot PDFComparison spectra PDF XIS0 spectrum from 20061022 showing CTI effects PDF 2007 Oct 1Attachments: Contamination rate plot PDF2007 May 7Here are the new parameters for the central contamination rate, as measured by E0102 from rev1.2 data, assuming C/O = 6, and where N_C = a*(1exp(cMJD/b)) with N_C = C column density in 10^18 cm^2. Note that I'm using MJD now instead of "days since 20050813". However I still assume zero contamination on 20050813, so (cMJD) = "days since 20050813" from the old form. Note that c is not a fitted value, it's just the mean observing time of the 20050813 data. ######################################################################## # as of Tue Apr 24 13:34:07 EDT 2007 # params are fitted through 20070410, using rev1.2, new model # new model = include HMXB RXJ0103, wilm abund, vern xsect # XIS0 define a 3.76558987299718 define b 282.597273120363 define c 53595.4083 # XIS1 define a 4.77791776540853 define b 197.744159380601 define c 53595.379 # XIS2 define a 4.07951865562546 define b 105.966875089484 define c 53595.3503 # XIS3 define a 5.99819756132834 define b 86.5701015652393 define c 53595.3542 Here are the old parameters which are used in the current xissimarfgen, just FYI: ######################################################################## # as of 20060825, using rev0.7, old model # these params are used in xissimarfgen contami file # ae_xi*_contami_20061016.fits # XIS0 define a 2.88939156841589 define b 170.437176167262 define c 53595.4083 # XIS1 define a 4.57935004724336 define b 176.836204404981 define c 53595.379 # XIS2 define a 3.95873438224372 define b 102.612617255168 define c 53595.3503 # XIS3 define a 5.88313373917308 define b 87.1269963157196 define c 53595.3542 I've also attached the latest figure, although I think maybe you already have it. The dotted line is the old fit, solid line is the new fit. Attachments: Contamination rate plot PDF2007 Apr 24Here is the latest contamination plot from the E0102 data. This includes all data through the 20070410 observation. Observations with SCI on have been included (and noted); these have been processed as per the web page instructions to remove the CTI correction. The processing is similar to the previous version of the plot. The solid lines are exponential fits to all the E0102 data points. The dotted lines show the exponential fits used in xissimarfgen (specifically the ae_xi*_contami_20061016.fits CALDB files which observers are currently using). This was based on rev0.7 processing and a slightly different spectral model. XIS1 and XIS3 are still pretty close, while XIS0 has deviated quite a bit from the projected model. I've included the PKS2155 crosscalibration results on the plot, but not in the fit. The XIS1 value is from the recent Sersic 15903 paper (Werner et al. 2007, astroph/0704.0475) while the others are from the Ishida et al. memo. I'm currently processing RXJ1856 rev1.x to include that, though if anyone has done this I'd be happy to use their data. The bottom panel shows the baseplate temperatures of each sensor, with the same color coding. XIS2 and XIS3 show a large excursion between MJD 54050 and 54100 (Dec 2006), presumably due to orbital effects. Intriguingly, the XIS3 contamination drops right after this. I'm not sure what to make of it. Attachments: Contamination rate plot PDF2007 Feb 27There are some differences in analysis between this plot and the previous one.
Attachments: Contamination rate plot PDFSpectral comparison PDF 2006 Oct 2Here is an EPS version of the contamination plot. I've extended and refit the exponential function to include the 20060825 data. Here are the new fit parameters, FYI: [N_C/1018 cm2] = a*[1exp(day/b)] # fit parameters # a units are 1018 cm2 # b units are days # day is number of days since 20050813 chip a b XIS0 2.889392 170.437176 XIS1 4.579350 176.836204 XIS2 3.958734 102.612617 XIS3 5.883134 87.126996 Attachments: Contamination rate plot:2006 Aug 4Here is a quick update on the contamination with the latest E0102 observation, dated 20060717. The FI chip values appear to be leveling off. I don't really understand the BI value, which is higher than before. I believe there are effects other than contamination that are driving this, and I'll try to summarize what I know for my talk on Tuesday at GSFC. The exponential fit for the BI ignores the 20050831 data point, otherwise the best fit is basically a straight line, which fits quite poorly. The error bars are 90%, and this still assumes C/O = 6. I'm in the process of estimating the A*Omega values for the latest points and some points in the future. If anyone else would like to do this as well for their favorite nonuniformity model, here are the central contamination fitting parameters: chip a t0  XIS0 3.01 183. XIS1 5.30 222. XIS2 3.99 105. XIS3 5.83 85.1 where: N_C = a*(1exp(t/t0)) with N_C = C column density in 1018 cm2 t = time in days Attachments: Contamination rate plot (PDF)Comparison of rev0.6 and rev0.7 (PDF) Contamination rate plot: 2006 Jul 12Here are some contamination updates. First, I've analyzed the latest (20060626) observation of E0102, and attached an updated contamination rate plot (first plot). The latest point is day 318, and it has not been included in the fit shown on the plot (dotted lines). The BI lowenergy effective area has decreased since the previous (20050521) observation, apparent from both the spectrum and the fitted contamination value. The FI effective area hasn't changed much, and is fairly consistent with extrapolation of the exponential function. Second, here are some numbers to use for the contamination uncertainty. I've run four additional models on the 20060416 data, and the table below shows the contaminant carbon column density that results. Two of the models (B & C below) bracket the assumed C/O ratio. As one would expect from the pureC results, the contamination carbon column increases with higher C/O ratio and vice versa. The other two models (D & E) adjust for possible energyindependent changes in the effective area, as perhaps seen in the Crab data. D assumes a 5% reduction, E 10%, and the fitted carbon column decreases. The A model is the "default" C/O = 6, fixed normalization version. The second and third tables list the difference and ratio (respectively) between each result and the default model. Some additional sources of error that you've probably thought of:
# E0102  contaminant N_C # DATE 20060416 # DAY 246.088 # rev0.7 processing # units are 10^18 C atoms cm^2 # A: C/O = 6, norm = 1 (default fit) # B: C/O = 3, norm = 1 # C: C/O = 12, norm = 1 # D: C/O = 6, norm = 0.95 # E: C/O = 6, norm = 0.90 XIS0 err XIS1 err XIS2 err XIS3 err  A 2.40 0.08 3.59 0.05 3.56 0.09 5.48 0.11 B 1.87 0.07 2.81 0.04 2.77 0.07 4.26 0.08 C 2.79 0.10 4.17 0.06 4.15 0.11 6.42 0.13 D 2.02 0.08 3.28 0.05 3.16 0.09 5.04 0.11 E 1.63 0.07 2.96 0.05 2.76 0.09 4.60 0.10 # residuals compared to default N_C # N_C(X)  N_C(A) XIS0 XIS1 XIS2 XIS3  B 0.53 0.78 0.79 1.22 C 0.39 0.59 0.59 0.94 D 0.38 0.31 0.40 0.44 E 0.77 0.62 0.80 0.88 # ratio compared to default N_C # N_C(X) / N_C(X) XIS0 XIS1 XIS2 XIS3  B 0.78 0.78 0.78 0.78 C 1.16 1.16 1.17 1.17 D 0.84 0.91 0.89 0.92 E 0.68 0.83 0.77 0.84 Attachments: Contamination rate plot (PDF)Comparison of rev0.6 and rev0.7 (PDF) Contamination rate plot: 2006 May 24I've constructed a new empirical model for the onaxis contamination evolution, this time assuming DEHP (C24H38O4, or C/O = 6 by number). Attached please find an updated history plot, as well as all fitted spectra for the two sources I used, E0102 and RXJ1856. Errorbars on the history plot are 90% confidence. For the fitting, I fixed C/O = 6, and only fit above 0.3 keV (0.31 keV for RXJ1856, 0.32 keV for E0102). I initially included a variable, energyindependent normalization factor in each fit, since the 12 keV region of E0102 shows a slight (~510%) countrate decrease which is difficult to explain from a simple C24O4 contaminant. The fitted normalization and contamination turned out to be highly correlated, however, so I abandoned this and froze the normalization for all chips and epochs for each object. The evolution model is a broken line, with the break set to the 20051216 observation. This seems to be the simplest (artificial) model given the sparse time samplings. Here are the parameters, first in column density of C and then in the xspec varabs C parameter (I'm no longer using "depth" in microns): # C column density # slope = 10^18 cm^2/day # inter = 10^18 cm^2 # break @ day 126.3 chip slope1 inter1 slope2 inter2 XIS0 0.012611 0.351729 0.008149 0.912850 XIS1 0.020215 0.009724 0.013369 0.794435 XIS2 0.025753 0.022866 0.007130 2.363714 XIS3 0.030836 1.029650 0.009726 3.682819 # C varabs parameter, assuming angr # slope = varabsC/day # inter = varabsC # break @ day 126.3 chip slope1 inter1 slope2 inter2 XIS0 0.003474 0.096895 0.002245 0.251474 XIS1 0.005569 0.002679 0.003683 0.218852 XIS2 0.007094 0.006299 0.001964 0.651161 XIS3 0.008495 0.283650 0.002679 1.014551 Note that varabsO = 0.07109*varabsC when using angr abundances. Some notes:
The XIS team is preparing an update to the spatialnonuniformity model, and Keith Arnaud will soon release a new xspec model akin to xisabs incorporating this model. Also, another E0102 observation has been completed and will be analyzed soon. Attachments: Contamination rate plot (PDF)E0102 spectra (PDF) RXJ1856 spectra (PDF) RXJ1856 Cband comparison (PDF) Contamination rate plot: 2006 May 10I'm posting the results of the most recent observation of E0102 (20060417). The XIS team has been discussing this via email, which is why this has taken a while, and discussions continue. I've attached a number of figures. Thanks to Hayashidasan, Mark, and Hamaguchisan for their help with this. Figure 1 shows the results of my initial fit to the new E0102 data, compared to the recent RXJ1856 and previous E0102 results and assuming pure carbon contamination. There is a clear discrepancy between E0102 and RXJ1856 at recent times. The errorbars for RXJ1856 reflect some discrepancies in the fits, based on (mostly) which energy range is included. This is still being discussed within the team. Figure 2 shows the equivalent C depth if DEHP is the contaminant (C24H38O4, so O/C = 1/6 by numer). Adding O has greatly reduced the necessary C in the E0102 fit, but hardly changed the C in the RXJ1856 fit. Note that this is only the C column density, the O also produces absorption, but is not represented directly on the plot. There is still some discrepancy between the columns derived from the two sources, but it is within the systematic error. It seems that the "C depth" inferred from E0102 is much more sensitive to the amount of O in the contaminant than the "C depth" inferred from RXJ1856. The reason for this can be seen in Figures 3 and 4. Figure 3 shows the count rate spectra of RXJ1856 (black) and E0102 (red) for the BI. Most of the flux of E0102 is above the O edge near 530 eV. Most of the flux of RXJ1856 is below this. Figure 4 shows the transmission of the two different contamination models used to fit the most recent E0102 point for XIS3 (the last blue point in Figures 1 and 2). The black line assumes C only (for the point in Figure 1), and the red line assumes O/C = 1/6 (for the point in Figure 2). The tranmission is identical above the O edge at 530 eV; therefore both of these contaminant models fit the E0102 data above 530 eV. Below this there is a discrepancy, but the count rate from E0102 is very low in this region. So in this source, absorption due to O is easily mimicked by pure C absorption, except when there are sufficient counts below 500 eV. (In the few obs with sufficient counts, O/C is constrained to be 1020%. In most of the obs, only a 90% upper limit of 20% is obtained.) RXJ1856 samples lower energies and the C edge well, and the transmission from O and C is obviously different here, so including O in the contaminant will not greatly change the amount of C. The last four figures show normalized count rates for three bright emission lines. I've also included values for RXJ1856 and Eta Carina, which have been scaled to the bestfit count rate for the initial observation. This is a modelfree way of showing the effective area changes, and a more consistent picture emerges. One might believe that the degradation is leveling off, especially for the worstaffected sensors XIS2 and XIS3. (Note that RXJ1856 and Eta Car count rates are for similar energy bands as the E0102 emission lines; they don't exhibit the lines themselves.) In summary, the discrepancy between E0102 and RXJ1856 (and probably other sources) might be due to the spectral regions being fitted and the assumed contamination composition. Incomplete knowledge of the response at these energies also complicates things, especially for the BI below 300 eV. The XIS team is working to produce the best contaminant model for users to employ in their spectral analysis. In the meantime, I (personally) have these caveats, which are open for discussion:
Attachments: PDF figures (8 pages)Figure 2 (rate plot): 2006 Feb 1Here is an analysis of the XIS contamination time dependence, as measured with four observations of 1E010272: obs. date exposure (ksec) 20050813 1.54.0 20050831 24 20051216 64 20060117 9.0 The contamination was measured as described in my previous XOOPS posting. This has been emailed to the SWG, but I include it here for completeness. A more thorough description of the analysis is forthcoming. The attached figure shows the bestfit contamination values for each epoch and detector (with errorbars), as well as linear fits for each detector. The t=0 data (20050813) were assumed to have no contamination; they were included in the fits anyway. For comparison, I've included the RXJ1856 results of Hayashidasan (filled circles). They line up reasonably well. The rates for XIS0,1,2 are consistent with being linear. XIS3 shows a fast increase during August, and a possible turnover at later times. Only carbon was included in this model. Including oxygen does not improve the fit, and amounts are consistent (within large errors) with zero. The oxygen contamination is estimated to be 10 % (+5%) of the carbon contamination based on the analysis of RXJ1856. However, this result is affected by the gain shift and needs further investigation for modeling. Note that these results and the empirical formula below are only valid for point sources observed at the XIS nominal aimpoint. The extraction region was the standard 6mm point source aperture assumed by the ARFs. CONTAMINATION TIME DEPENDENCE The best fit lines have the following parameters plus errors, with zero time defined as 20050813 (177242000 seconds since 20000101, MJD 53595.412): change in effective carbon column density XIS0 XIS1 XIS2 XIS3 slope (1e16 cm^2/day) 1.6 +0.1 2.7 +0.1 3.1 +0.1 4.1 +0.5 intercept (1e16 cm^2) 4.4 +4.0 9.6 +15 3.2 +14 54. +50. The easiest way to include this absorption component within XSPEC is to use the "varabs" model. This enables one to set the H column to zero, which cannot be done completely with the "vphabs" model, and also have only one parameter for each absorbing element. See the XSPEC help for more information. The following time models show the change in the varabs(3) parameter, assuming the "angr" abundance table (C/H = 3.63e4) and the same zero time as above: change in varabs carbon parameter XIS0 XIS1 XIS2 XIS3 slope (varabsC/day) 0.0043 0.0073 0.0085 0.0114 intercept (varabsC) 0.012 0.026 0.009 0.148 The time2day and time2sec ftools (and friends) are helpful in determining the time offset of a particular observation. Attachments: PDF file2006 Jan 24Here are preliminary results for the latest E0102 observation. I've fitted spectra for XIS1 and XIS3 (see attached). Here are the numbers for the contamination, which has increased (in fitted absorption depth) by about 15%, a difference of about 2 sigma (fit parameter error). changes from 20051216 to 20060117: XIS1 XIS3 C depth: +17% +13% gain shift: 4.3ev 1.4ev I've included my whole original table, with the XIS1 and XIS3 numbers updated. The spectra show a pretty obvious decline in the O line intensities, anyway, as well as the low energy continuum. This is more obvious in the BI than the FI.  surface density of carbon (10^5 grams/cm2) date XIS0 XIS1 XIS2 XIS3 20050813 0 0 0 0 20050831 0.9 +0.2 0.2 +0.2 0.5 +0.3 1.9 +0.2 20051216 2.6 +0.3 6.4 +0.5 7.6 +0.4 10.4 +0.6 20061217 7.4 +0.4 11.6 +0.6  thickness (microns; assumes density of 2.2 g/cm3) date XIS0 XIS1 XIS2 XIS3 20050813 0 0 0 0 20050831 0.04 +0.01 0.01 +0.01 0.02 +0.01 0.08 +0.01 20051216 0.12 +0.01 0.29 +0.02 0.35 +0.02 0.47 +0.03 20061217 0.34 +0.02 0.53 +0.03  gain shift (eV) date XIS0 XIS1 XIS2 XIS3 20050813 6.3 +1.0 1.7 +0.7 6.8 +0.8 5.1 +0.9 20050831 11.4 +0.4 11.5 +0.3 11.2 +0.4 10.9 +0.4 20051216 12.7 +0.4 15.4 +0.2 14.4 +0.4 13.4 +0.4 20061217 19.7 +0.4 14.8 +0.7 Attachments: PDF file of all spectra2006 Jan 18Here is a first analysis of the three observations of 1E010272.3, in an attempt to characterize the contamination. I have attached spectra showing the model fit for the 20050813, 20050831, and 20051216 observations for each detector. Here are some measures of the contamination and gain "shift", with caveats as mentioned below. Fitting errors are all in the 10% range, although systematic errors are likely larger and less welldetermined.  surface density of carbon (10^5 grams/cm^2) date XIS0 XIS1 XIS2 XIS3 20050813 0 0 0 0 20050831 0.9 +0.2 0.2 +0.2 0.5 +0.3 1.9 +0.2 20051216 2.6 +0.3 6.4 +0.5 7.6 +0.4 10.4 +0.6  thickness (microns; assumes density of 2.2 g/cm^3) date XIS0 XIS1 XIS2 XIS3 20050813 0 0 0 0 20050831 0.04 +0.01 0.01 +0.01 0.02 +0.01 0.08 +0.01 20051216 0.12 +0.01 0.29 +0.02 0.35 +0.02 0.47 +0.03  gain shift (eV) date XIS0 XIS1 XIS2 XIS3 20050813 6.3 +1.0 1.7 +0.7 6.8 +0.8 5.1 +0.9 20050831 11.4 +0.4 11.5 +0.3 11.2 +0.4 10.9 +0.4 20051216 12.7 +0.4 15.4 +0.2 14.4 +0.4 13.4 +0.4 These numbers and the attached spectra seem to show the contamination is getting worse for most of the chips, compared to the RXJ1856 observations in October. XIS0 seems to have changed little. Also, there's evidence for contamination as far back as August, akin to what Takeisan sees in his first post. The model, provided by Paul Plucinsky of CfA, is a bremsstrahlung continuum with 24 Gaussian emission lines culled from Chandra and XMM grating results and adapted (by Paul) to fit the ACIS data and track changes in the response. There are two absorption components, one wellconstrained Galactic component and one poorlyconstrained SMC component. The SMC column is probably the major systematic error. I have also included a varabs component with only carbon for the contamination. For each XIS detector, I have fit the model to all three observations simultaneously. The contamination absorption column of the 0813 spectrum was fixed at zero, the other two were allowed to float independently. I fixed the emission line centers, but needed a width (sigma) of about 15 eV to help the spectral resolution come out right. There appear to be some residual problems with the line widths in the spectra, in any event. Finally, I allowed the line and continuum normalizations to vary (tied between the observations), as the fit was poor using the straight ACIS values. The fitted values are generally within 10% of the ACIS model numbers. Thus I have effectively used the (short) 0813 observations as a zerocontamination baseline for the later two observations. The chip temperature was a bit higher for this dataset (80 vs. 90 C), but ground tests show this should have little impact on the QE of the FI chips (2% or less). The BI chip is a different story, but probably the QE at 80 C is within 20% of the 90 C value near 0.5 keV (although there's very little information to back up that statement). Attachments: PDF file of all spectraPresentationsClick on a preview image to download the full PDF presentation file.
